Jump to Content

Generally Available: C3 VMs with 4th Gen Intel Xeon and industry-leading price-performance

May 22, 2023
Salil Suri

Director, Product Management, GCE

Foster Casterline

Product Manager, Google Compute Engine

Editor's noteUpdate, July 31, 2023: C3-standard and highmem machine types are now Generally Available. C3-standard with Local SSD is now in Public Preview. 

Today, we are excited to announce general availability of the C3 machine series, bringing the enterprise-grade performance and reliability of the 4th Gen Intel Xeon scalable processor (code named Sapphire Rapids) to Google Compute Engine (GCE) and Google Kubernetes Engine (GKE) customers — all with industry-leading price-performance.  

The C3 machine series is a general-purpose VM that combines the latest innovations in compute, networking, and storage into one platform. This enables C3 to deliver price-performance improvements for a wide variety of workloads. Based on our benchmarking, compared to our previous generation C2, web serving applications such as NGINX can see up to an 88% improvement in price-performance, relational databases such as Postgres or MySQL up to 40%, in-memory databases such as Redis up to 39%, and data analytics such as Hadoop up to 18%. When benchmarked against comparable offerings from other cloud providers as of May ‘23, C3 provided an average of up to 25% better price-performance.


Built on the latest-generation 4th Gen Intel Xeon and DDR5, C3 VM shapes are optimized for the underlying NUMA architecture to deliver consistent performance and can scale up to 176 vCPUs. C3 VMs use DDR5 memory, which is 50% faster than DDR4. C3 is currently available in the highcpu (2GB/vCPU) memory configuration. Standard (4GB/vCPU) and highmem (8GB/vCPU) configurations, as well as local SSD (up to 12TB) will be available in the coming months.

The new C3 machine family is a major leap in our purpose-built infrastructure offerings:

  • Using the Infrastructure Processing Unit (IPU) that we co-developed with Intel to offload networking

  • Delivering high performance block-storage with Hyperdisk

  • Speeding up ML training and inference with Intel AMX

Let’s take a look at some innovations you’ll find in the C3 family.

Latest generation networking with the IPU

At Google Cloud Next ‘22, we announced Google’s and Intel’s custom co-developed Infrastructure Processing Unit (IPU), which powers C3 and our new machine series moving forward. Google’s IPU includes an offloaded Andromeda virtual network stack and programmable packet processing engine that enables higher levels of networking performance, isolation and security. The offloaded networking frees up CPUs allowing C3 to offer larger VM shapes with less interference to customer workloads and applications from networking I/O. This also enables C3 VMs to deliver up to 200 Gbps low-latency networking — 2x our previous generation VMs, making it ideal for tightly-coupled distributed computing, high-performance computing (HPC) and network-intensive workloads.

Latest generation storage with Hyperdisk

Hyperdisk is the latest generation block storage service in Google Cloud. Hyperdisk leverages our IPU technology to deliver significantly higher levels of performance, flexibility, and efficiency by decoupling storage processing from virtual machines. With Hyperdisk, you can dynamically scale storage performance and capacity independently to efficiently meet the storage I/O needs for data-intensive workloads, such as data analytics and databases. Now, you don’t have to choose expensive, larger compute instances just to get the higher storage performance. On C3, Hyperdisk block storage delivers up to 4x higher storage throughput and 10x higher IOPS over our previous-generation C2 VMs. 

Intel AMX for AI/ML

The 4th Gen Intel Xeon processor comes with Intel Advanced Matrix Extensions (AMX), a built-in accelerator that significantly improves the performance of deep-learning training and inference on the CPU. For example, when AMX is enabled, BERT (a machine learning framework for natural language processing) matrix multiplication using the bfloat16 format, throughput improved up to 12x when compared to AVX512 to perform the same operations. For similar sized matrices using the INT8 format, throughput improved up to 2x when compared to AVX512. The AMX instructions are integrated into the latest PyTorch and TensorFlow frameworks so customers can easily leverage the capability of this built-in accelerator.

“Businesses now more than ever need a holistic AI strategy that’s uniquely designed for performance, productivity and efficiency to meet the compute demands of today and tomorrow. Google Cloud and Intel have been co-innovating together since 2016 and our work continues to focus on addressing the growing need for democratized AI across the industry. The built-in capabilities we provide with our 4th Gen Intel Xeon Scalable processors unlock valuable AI performance per watt workload acceleration without any code changes.” – Kavitha Prasad, VP and GM, Data Center, AI, Cloud & Enterprise Strategy (DACES), Intel Corporation

Enhanced maintenance experience

C3 continues to raise the bar on seamless maintenance, leveraging a variety of technologies that allow Google to update software and firmware without disrupting customer workloads while keeping the infrastructure updated and secure. These features make C3 ideal for running your most demanding business-critical workloads.

What are our customers are saying


"At Palo Alto Networks, we develop and deploy deep learning models for inline threat detection in our customers' network traffic. Inference latency is critical for our AI workloads. By adopting C3 VMs with Intel Sapphire Rapids and the new AMX instruction set for AI, we are seeing 2x performance for some of our inline models, compared to the previous generation N2 Ice Lake VMs." - Suiqiang Deng, Sr. Distinguished Engineer, Palo Alto Networks


"With the new C3 machine series, we've noticed artists enjoy a 30% performance increase, making it easier and faster to bring their creative visions to life. We've observed substantial improvements in both 3D and 2D rendering across all of our clients, making C3’s performance level a necessity for leaders in the VFX industry." - Tom Taylor, Founder, Gunpowder Tech LLC


At Cadence, we're excited that initial testing indicates a 15% reduction in simulation runtime per core for Clarity workloads running on C3 vs C2. These significantly faster runtimes on Google's Cloud Platform will help to increase engineering productivity for our joint customers.” - Ben Gu, Vice President of R&D, Cadence

Get started with C3 today

C3 VMs are available today in the following regions: us-central1 (Iowa), us-east1 (S. Carolina), us-east4 (North Virginia), europe-west1 (Belgium), europe-west4 (Netherlands), and asia-southeast1 (Singapore). To start using C3 instances, select C3 under the General Purpose machine family when creating a new VM or GKE node pool in the Cloud console. Stay up-to-date on regional availability by visiting our regions and zones page or contact your Google Cloud sales representative for more information.

Posted in